Abstract
With the explosion amount of information caused by Internet expansion, machine learning is getting more and more popular recently. However, real world application of machine learning theory is still very challenging and hard to achieve. One of the main difficulties stops the direct application is the issue called data loss. No matter data loss is caused by sensors themselves or storage leakages, once the some data are lost, it is extremely hard to recover them. The probability of data loss is incredibly high and common in real world. For example, a record file of oral speech online might have data loss because of the following reasons: 1) microphone receives too much noise from the receiver endpoint, 2) timeworn recording device misses some audio information when writing into disks, or 3) the audio information is lossy-compressed when it is uploaded. Once the file loss has been made, there would be error in when model is built up and when model is making predictions. Thus, it is very crucial for me to derive an effective but robust algorithm that can impute missing information and reach better classification results.
Keywords
Real-time Algorithm, Impute Missing Data, Iterative Learning